training/evaluate.py

"""
Evaluation harness for comparing agents.

Runs multiple episodes per agent and computes comprehensive metrics.
"""

import sys
import os
import io

# Fix Windows console encoding for Unicode output
if sys.stdout.encoding != "utf-8":
    sys.stdout = io.TextIOWrapper(sys.stdout.buffer, encoding="utf-8", errors="replace")
    sys.stderr = io.TextIOWrapper(sys.stderr.buffer, encoding="utf-8", errors="replace")

sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))

from env.assistant_env import ExecutiveAssistantEnv
from agents.random_agent import RandomAgent
from agents.rule_based_agent import RuleBasedAgent
from agents.rl_agent import RLAgent


def evaluate_agent(env, agent, num_episodes=50, label="Agent"):
    """Evaluate an agent over multiple episodes.

    Returns:
        Dict with aggregated metrics.
    """
    all_rewards = []
    all_metrics = []
    all_steps = []

    for ep in range(num_episodes):
        state = env.reset()
        total_reward = 0.0
        done = False
        steps = 0

        while not done:
            action = agent.act(state)
            state, reward, done, info = env.step(action)
            total_reward += reward
            steps += 1

        all_rewards.append(total_reward)
        all_steps.append(steps)

        metrics = info.get("metrics", {})
        all_metrics.append(metrics)

    # Aggregate
    avg_reward = sum(all_rewards) / len(all_rewards)
    avg_steps = sum(all_steps) / len(all_steps)

    avg_metrics = {}
    if all_metrics:
        for key in all_metrics[0]:
            values = [m.get(key, 0) for m in all_metrics]
            avg_metrics[key] = sum(values) / len(values)

    return {
        "label": label,
        "avg_reward": round(avg_reward, 2),
        "avg_steps": round(avg_steps, 1),
        "best_reward": round(max(all_rewards), 2),
        "worst_reward": round(min(all_rewards), 2),
        "metrics": {k: round(v, 3) for k, v in avg_metrics.items()},
    }


def main():
    """Run full evaluation comparison."""
    NUM_EVAL = 100
    os.makedirs("logs", exist_ok=True)

    print("=" * 60)
    print("📊 AI Executive Assistant — Agent Evaluation")
    print("=" * 60)

    agents = {
        "Random": RandomAgent(seed=123),
        "Rule-Based": RuleBasedAgent(),
        "Q-Learning (untrained)": RLAgent(epsilon=0.0, seed=123),
    }

    results = []
    for name, agent in agents.items():
        print(f"\n  Evaluating {name}...")
        env = ExecutiveAssistantEnv(difficulty="medium", seed=123)
        result = evaluate_agent(env, agent, NUM_EVAL, label=name)
        results.append(result)

    # Print results table
    print("\n" + "=" * 60)
    print("📈 EVALUATION RESULTS")
    print("=" * 60)
    print(f"{'Agent':<25} {'Avg Reward':>10} {'Completion':>12} {'Efficiency':>12}")
    print("-" * 60)

    for r in results:
        m = r["metrics"]
        print(
            f"{r['label']:<25} "
            f"{r['avg_reward']:>10.1f} "
            f"{m.get('task_completion_rate', 0):>11.1%} "
            f"{m.get('efficiency_score', 0):>11.1f}"
        )

    print("\n" + "=" * 60)


if __name__ == "__main__":
    main()